reddit · SideProject · 10-03 18:50 76 分 ← 返回列表

I got tired of wondering where all my Claude Code tokens were going, so I built optimAIzr

I built optimAIzr because I wanted to see where I was wasting tokens and money with Claude Code/Codex. ~2 weeks later: 📦 2k+ npm downloads 🌐 1.8k+ website visitors It is fully locally, it finds oversized context, model mismatches, error loops, and other usage that could probably be cheaper. I ran it on my own sessions and analyzed ~1.25B tokens. Around 15% looked potentially optimizable. It’s open source now, with live recommendations inside Claude Code. Next up: the hosted dashb

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：Claude Code/Codex 用户无法看清 token 与费用的浪费点，需要本地分析上下文过大、模型错配、错误循环等可优化项。

目标用户：高频使用 Claude Code/Codex 的独立开发者、小型 AI 工程团队和需控制 LLM 成本的开发者。

证据：I built optimAIzr because I wanted to see where I was wasting tokens and money with Claude Code/Codex. ~2 weeks later: 📦 2k+ npm downloads 🌐 1.8k+ website visitors. I ran it on my own sessions and analyzed ~1.25B tokens. Around 15% looked potentially optimizable. Next up: the hosted dashboard

变现：开源 CLI 引流，托管仪表盘按席位/月订阅收费；团队版提供成本归因、预算告警、优化建议和报表；企业版支持私有部署与 SSO。

MVP：2 周内做本地 CLI：解析 Claude Code/Codex 会话日志，统计 token/成本，标记超大上下文、模型错配、错误循环，输出可执行优化建议；再加一个极简 Web 仪表盘上传摘要并生成周报。

竞品：ccusage、Claude Code Usage Monitor、Helicone、LangSmith、OpenLLMetry

摘要：作者因想弄清 Claude Code/Codex 的 token 和费用浪费，做了本地开源工具 optimAIzr，2 周获得 2k+ npm 下载和 1.8k+ 网站访问。自测分析约 12.5 亿 tokens，发现约 15% 可优化，下一步计划推出托管仪表盘。该需求真实且有增长信号，但尚未看到明确付费或收入数据。